基于低秩矩阵填充的推荐算法

针对已有协同过滤推荐技术中评分矩阵极度稀疏问题,提出了一种基于低秩矩阵填充技术的推荐算法.该算法从贝叶斯框架出发,提出了能够解决低秩矩阵问题的分层高斯先验模型,并将广义近似消息传递算法嵌入到贝叶斯框架,规避了贝叶斯学习过程中烦琐的矩阵逆运算,提升了算法运算速度,同时在广义近似消息传递算法中施加阻尼运算以促进收敛.在开放数据集上的实验结果表明,所提出的算法与相关的矩阵填充推荐算法相比,有效地提高了推荐准确度.

Recommendation Algorithm Based on Low-Rank Matrix Completion

The existing collaborative filtering recommendation algorithm has the defect of sparse score matrix. To solve this problem, a recommendation algorithm based on low-rank matrix completion is proposed. By applying the variational Bayesian framework, a hierarchical Gaussian prior model was adopted to encourage a low-rank solution. To avoid cumbersome matrix inverse operations and improve computing speed, the generalized approximate message passing technique was used and embedded in the variational Bayesian framework. Meanwhile, the damping is introduced into the algorithm to promote convergence. The experiments of open datasets show that the proposed method can achieve better prediction accuracy compared with the related matrix completion algorithm.